February 9, 2009
There's nothing like hearing the results of studies directly from those who actually conducted the research. In this interview, you'll meet two of these impressive HIV researchers and read their explanation of studies they presented at CROI 2009. After their explanation, they will answer several questions from the audience. This discussion was moderated by John Mellors, M.D.
Mari Kitahata: We've been struggling with this question of when to initiate HAART [highly active antiretroviral therapy] in asymptomatic HIV-infected patients for some time -- largely because the methodology is quite complex and the sample size required to address the question requires large numbers of patients over a long period of time.
Mari Kitahata, M.D.
As you know, recently updated U.S. guidelines recommend treating all asymptomatic patients with a CD4 count less than 350. In a previous study, we examined patients who had either initiated or deferred HAART at a CD4 count between 351 and 500 and found a significant improvement in survival. Emerging data about the continued potential harm of ongoing chronic inflammation and HIV replication, as well as potential persistent and permanent immunologic defects, raise the [idea that therapy] even earlier in HIV disease, at a CD4 count greater than 500, could significantly improve survival.
To address some of the major challenges, we used more advanced statistical methods that allow for mimicking a clinical trial using observational data. We also combined data from all of the major HIV cohorts in North America throughout the U.S. and Canada [into the North American AIDS Cohort Collaboration on Research and Design (NA-ACCORD)] and examined patients who were enrolled at the same CD4 cell count level above 500 between 1996 and 2006.1 Nine thousand one hundred fifty-five patients were enrolled in the study. They all had a CD4 count greater than 500. They had never received antiretroviral treatment and were AIDS free. Twenty-nine percent, or 2,616, began taking HAART at a CD4 count greater than 500, and the remaining 71% deferred. After controlling for potential confounding factors, we found a 60% higher risk of mortality for patients who deferred treatment at a CD4 count greater than 500. Our findings suggest that therapy earlier in HIV infection could significantly prolong survival.
We designed our study to address the effect of HAART on all-cause mortality. I think the growing appreciation of the importance of non-AIDS conditions, such as liver, cardiovascular and renal disease -- not previously thought to be related to HIV -- make it important not to use simply a combined endpoint of a composite time to AIDS event or death. So we designed our study to look at all-cause mortality.
The next step in our analyses will be to look at causes of mortality and collect information in more detail on the non-AIDS-related serious conditions. Our plans are to do modeling that involves looking at competing risks of those events and the earlier initiation of HAART in the North American cohort.
Jonathan Sterne, M.D.
Jonathan Sterne: Our study also addressed the question of the CD4 cell count at which it's beneficial, for either rates of AIDS and death or for mortality rates, to start treatment with combination therapy.2 We analyzed data from cohorts contributing to the ART Cohort Collaboration. It's a collaboration that I lead that has produced a number of publications looking at the prognosis of HIV-infected patients from the time that they started therapy.
As Mari said, the statistical analysis of this issue in observational data is complicated. What you'd like to do is to mimic a randomized trial that randomizes people to start at a higher CD4 threshold or defer to a lower CD4 range. From the data in the ART Cohort Collaboration, we followed people from the start of therapy, so we're missing the data you would have in an RCT [randomized controlled trial]. In the people who deferred, there are two possibilities: They could either progress to AIDS or die without ever starting treatment, or they could have some time off treatment before they started. We've accounted for those things by using data on 21,000 people from seven cohort studies that followed people in the pre-HAART era from 1989 to 1995, before effective combination therapy became available.
We looked at different thresholds for starting HAART, beginning in the lower range -- for example, by comparing a strategy of initiating in the range of 200 to 100 [cells/mL], compared with deferring to the range of 100 to 0. Then we increased the threshold in increments of 25 until we compared initiating in the range of 550 to 450 with deferring to the range of 450 to 350.
The first thing to say about our findings is that they clearly demonstrate the substantial adverse effects of delayed treatment in the lower CD4 ranges. They show that delaying to below 250 cells/mL is clearly associated with increased rates of AIDS and death as a combined endpoint, and of mortality. That really emphasizes the need to diagnose HIV infection early enough to get people on treatment when they can benefit most from it.
We found some evidence that starting at above 350 cells/mL, as recommended following recent changes to guidelines, is beneficial compared to deferring to the range of 350 to 250, with a hazard ratio of 1.28. That's for rates of AIDS and death -- effects on mortality were less dramatic.
We didn't find benefit from starting [treatment] at above ranges of approximately 400. Our hazard ratios are somewhat more conservative, close to 1.0, than those estimated in the NA-ACCORD study.
John Mellors: Can I ask what was the duration of follow-up after initiation in both studies?
Jonathan Sterne: We restricted to patients who'd started therapy after 1998 on the basis that combination therapy was less effective in 1996 to 1997 than subsequently. I think the median -- I have to look it up -- is about three years, with follow-up of up to about six years.
Mari Kitahata: I want to point out the strength of the two different methods that we used to reach somewhat similar findings. A difference in our study is that we started everyone at the same starting point, at greater than 500, so it was a head-to-head comparison that didn't involve the lead-time bias issue. It had 28,000 person-years of follow-up. The method we applied to be able to mimic a randomized trial included informative censoring, and therefore the median time is somewhat altered by including the time up to censoring, but it was similar: approximately 3.5 years, and up to eight years in some patients.
John Mellors: I remember several years ago, when initial data were presented in short-term follow-up looking at AIDS or death, that the field tended to move towards reaching the edge of the precipice and see how close you could get without showing a signal that there was a greater risk of a deferral strategy. Now I sense, and I think it's the right thing, that we're going to be moving away from that and we have moved away from that -- get to the precipice and then start honing in on higher and higher CD4 starting points.
With that, why don't we open it up for questions?
Reporter #1: You said that you looked at all-cause mortality, but did you see any other benefits or harm with starting treatment earlier, other than intuitive, quality-of-life things?
Mari Kitahata: I think you're raising the issue of important factors about patients in routine care that may be different from more selected patients in clinical trials. I want to emphasize -- and this gets also to the question of treatment -- that the NA-ACCORD encompasses almost 100,000 patients overall, and the greater-than-500 start period, at least in the U.S., was in the hit-hard, hit-early David Ho era. The peak of our initiation of therapies was with the older therapies -- actually, the peak was only 16% of a given set of patients within a year initiating at greater than 500 [cells/mL], and that was in 1998. So we're looking at a majority of people who were treated with non-boosted PIs [protease inhibitors], with the second group being on NNRTI [non-nucleoside reverse transcriptase inhibitor]-based regimens. Importantly, we looked at all comers, which included people who had injection drug use [IDU], hepatitis C virus [HCV] infection and other potential factors that might be associated with increased mortality, as well as potentially the exposure group. Our results were robust when controlling for those things. We were not addressing specific issues of other outcomes. We are collecting data now on non-AIDS morbidity and mortality, and those will be important things to examine in the future studies.
Reporter #2: The new guidelines that came out last August suggested that 350 was a good place to start, but that you could start at above 500 if you wanted. I'm wondering if the data that you're presenting, in your view, should alter guidelines at all.
Mari Kitahata: Our previous study that was presented at ICAAC [Interscience Conference on Antimicrobial Agents and Chemotherapy], which is hopefully going to be published shortly, demonstrated a significant improvement in survival starting at a CD4 count of 351 to 500. The DHHS [U.S. Department of Health and Human Services] guidelines and the IAS-USA [International AIDS Society-USA] guidelines both make 350 the threshold cutoff below which all asymptomatic patients should start therapy, but make exceptions for other subgroups of patients -- perhaps those with renal disease, hepatitis and even cardiovascular disease. There are indications that those subgroups of patients perhaps would benefit even more with earlier therapy.
What I can tell you is that this study shows that [starting therapy above] a threshold of 500 -- which, remember, is not near normal -- is associated with significant improvement overall, over the 10-year period of the study. Patients with other comorbid conditions, such as hepatitis C virus infection, benefited from the earlier treatment perhaps even more. And there was still an independent effect on mortality for having those conditions, above and beyond the benefit of starting HAART earlier.
Jonathan Sterne: I think our answers on this point diverge somewhat. In the analyses that I'm presenting, we did see evidence of benefit from starting at above 350, but if we go up the threshold, as high as 450, we didn't see evidence of further reductions in rates of AIDS and death or mortality.
I think it's important to emphasize: For our analysis, and possibly for the NA-ACCORD analysis as well, these are analyses of observational data, and there is always the possibility that the analyses are affected by confounding. If beyond what we've measured in the data, there are factors that are associated with the patient's prognosis that lead their doctors to postpone treatment, then that could bias the results and make the benefit from starting earlier look better than it really is.
The final point I'd make is that, starting above 350, absolute rates of mortality, and absolute rates of AIDS and death combined, are actually relatively low. The higher you put your threshold, the smaller the absolute benefit in terms of, say, your three-year risk of dying. And when absolute benefits are small, you then have to think about balancing the benefit against disadvantages, such as treatment and toxicity.
Mari Kitahata: I'll just make two further comments. We did two additional analyses addressing that issue. One: Although these are complex progression analyses, we did do an estimated, calculated, cumulative mortality rate that demonstrated that the absolute risk of death for patients that were deferring at a CD4 count greater than 500 was significantly higher throughout the 10-year period of the study.
The other analysis we did was a Monte Carlo simulation to address the issue of unmeasured confounding. After controlling for calendar year, cohort, sex, age, race, history of IDU, HCV infection, CD4 and viral load, we looked at whether an unmeasured confounder could have mitigated our results. What we found was that a confounder would have to have an association with relative hazard of mortality of 4.0 and an odds ratio associated with deferral of treatment of 4.0 to reduce the relative hazard of our results from a mortality risk of deferral of 1.6 to only 1.3. So an unmeasured confounder would have to be extremely large to mitigate the results that we found.
John Mellors: How feasible would it be to have a randomized clinical trial of deferred versus immediate therapy in individuals at whatever threshold you want?
Mari Kitahata: I think the points Jonathan is making about observational data are very important. That's why we went the extra mile to really try to do all the things that we could do to address potential confounding. We feel that we've applied those methods and had robust results in all sensitivity analyses we've done. The advantage of observational data is that we have been able to analyze data on a cohort of 100,000 patients overall, over a 10-year period between 1996 and 2006, in a collaboration that has only begun.
I think any study that will try to address this question will take large numbers of patients and long periods of time. And that is one of the advantages of prospectively collected cohort data. I think a very large randomized trial would be needed to balance all unmeasured factors between treatment and control groups, but certainly, observational data has the potential for confounding to remain.
Jonathan Sterne: The question is, "Is it feasible?" My colleagues who were involved in proposing such a trial would tell me that it absolutely is feasible.
If you believe it's feasible, then I think the question is, "Should it be done?" I would say that [decision is] slightly above my pay grade, but you got to balance the potential cost of potentially tens of millions of dollars of doing such a trial against the fact that only by doing a trial, and a well-conducted trial, can you definitively answer the question in terms of excluding confounding by factors that you don't measure in the cohorts.
I'm still very struck by the fact that the SMART trial, which as I understand it was also a trial that was hotly debated before it happened -- the trial of treatment interruption versus continuous treatment -- produced results that were unexpected at the time the trial was conceived, and that have really informed the way we understand the benefits of therapy since then.
Reporter #3: Mari, you said that 500, of course, is not normal. But we also know that normal CD4 counts tend to vary by racial/ethnic groups and things like that. Have you done any analysis along those lines, and might there be some implication -- if we're going to start at higher CD4 counts -- whether those might be better guided or framed along racial/ethnic lines and what is normal for those particular patient populations?
Mari Kitahata: Yes, it's a complex question. What I can say from our data is that race is a difficult variable, and there's misclassification because it's self-reported. The classifications used in Canada are a bit different from the U.S. And they're never perfect measures, I think, of what they're trying to measure.
But given those caveats, we did include the major categories of white race, black race, Hispanic and other, and did not find an effect of race on our results. There are many ways in which those kinds of factors can impact patient outcomes, but it was not a major factor in our study.
Reporter #4: I'm wondering how you reconcile your data with other data that is continuing to be discussed at meetings like this regarding risk for cardiovascular side effects and metabolic side effects in relation to all-cause death and survival. ... How do we reconcile this data with other data that is still pushing for limiting drug selections, when even the most dangerous drugs in the earlier days give people a better survival advantage?
Jonathan Sterne: We were talking about some of these issues over breakfast this morning. And one of the points that came up in that conversation is that the real problem for the majority of patients is that they are not diagnosed and therefore not treated until their CD4 counts are well below even what the current guidelines say. As Mari says, even in, what was it, 1998, at the peak of "treat hard, treat early," it was only 16% of patients [on treatment]. And the main reason that patients in rich countries are not treated is that they're just not diagnosed.
If you're worried about the public health impact, then I think Mari and I, and pretty much everyone else, can agree that if you get a patient with 250 cells, they should be treated. Why are they not treated? Primarily, it's because they haven't yet been diagnosed. In public health terms, I think that raises issues of, "Should we be looking harder for undiagnosed infection?"
Mari Kitahata: I think the dramatic improvement in survival since the advent of HAART has made HIV a chronic disease -- in resource-rich countries, at least -- and this is due to a dramatic decline in AIDS-defining illnesses. But as we're appreciating more and more, there are also HIV-related contributions to these classically non-AIDS-related conditions; there are several recent reviews of this. Judy Aberg did one that was quite informative about cardiovascular risk.
I think, again, it's a very complex issue. And clearly, as Jonathan pointed out, SMART informed us that some of these "non-AIDS conditions" are in fact significantly contributed to by HIV infection and had declined in the HAART era, and that the cardiovascular risks and toxicities -- which are a hot area of investigation and need much, much further work -- are actually modest compared with the benefits of HAART.
This transcript has been lightly edited for clarity.